Google DeepMind Unveils TxGemma: New Open AI for Drug Discovery

Originally published at ssojet Google DeepMind has launched TxGemma, an open collection of AI models aimed at improving drug discovery and clinical trial predictions. Built on the Gemma model family, TxGemma is designed to streamline the drug development process and accelerate the identification of new treatments. The TxGemma model addresses the significant challenges of the drug discovery process, where 90% of drug candidates fail to progress beyond phase 1 trials. By utilizing large language models (LLMs), TxGemma enhances the prediction of therapeutic properties throughout the research pipeline. Researchers can identify promising drug targets and assess clinical trial outcomes more effectively. Jeremy Prasetyo, CEO of TRUSTBYTES, mentioned, "AI that explains its own predictions is a game-changer for drug discovery—faster insights mean faster breakthroughs in patient care." TxGemma, as a successor to Tx-LLM, expands its capabilities and is available as an open-source model. Image courtesy of Blockchain.News The model is trained on 7 million examples and comes in three sizes—2B, 9B, and 27B parameters. It specializes in: Classification: Predicting molecule permeability across the blood-brain barrier. Regression: Estimating drug binding affinity. Generation: Inferring reactants from chemical reactions. In benchmark tests, the 27B Predict model outperformed or matched specialized models on 64 of 66 key tasks. For further details, refer to the published paper. TxGemma-Chat allows for interactive AI experiences, enabling researchers to ask complex questions and receive detailed explanations. Researchers can fine-tune the model for specific needs using a Colab notebook. Additionally, Google DeepMind is introducing Agentic-Tx, which integrates TxGemma into multi-step research workflows. This tool combines TxGemma with Gemini 2.0 Pro and utilizes 18 specialized tools to enhance research capabilities, as demonstrated in benchmarks like Humanity’s Last Exam and ChemBench. TxGemma is available on Vertex AI Model Garden and Hugging Face, which allows researchers to experiment with the models and use fine-tuning tools. Introduction of Gemma 3 The Gemma family has expanded with the introduction of Gemma 3, which is described as the most capable, portable, and responsible open model yet. Celebrating over 100 million downloads, the Gemma models have fostered a vibrant community with over 60,000 variants. Gemma 3 is designed for high performance on single GPU or TPU hosts, with several key capabilities: Outperforming competitors like Llama3-405B and DeepSeek-V3, it provides state-of-the-art performance for its size. Supporting over 35 languages out-of-the-box, with pre-trained support for up to 140 languages. Enhanced text and visual reasoning capabilities for applications analyzing images, text, and videos. A 128k-token context window for processing large volumes of information. Function calling capabilities to automate tasks and enhance agentic experiences. Developers can seamlessly integrate Gemma 3 into their existing workflows using tools like Hugging Face Transformers, Ollama, and Google AI Edge. Gemma 3 and its safety counterpart, ShieldGemma 2, offer built-in safety measures for image applications, categorizing outputs across three safety areas: dangerous content, sexually explicit content, and violence. Gemini 2.5 Model Launch Gemini 2.5, described as Google's most intelligent AI model, enhances reasoning capabilities. The first release, Gemini 2.5 Pro Experimental, leads benchmarks significantly, showcasing strong reasoning and coding abilities. Image courtesy of Google AI Gemini 2.5 models can analyze information, draw logical conclusions, and make informed decisions. The new model features: Advanced reasoning capabilities that surpass previous versions. A 1 million token context window, with plans for a 2 million token window. Enhanced performance across math, coding, and science benchmarks. Developers can start experimenting with Gemini 2.5 Pro in Google AI Studio or through the Gemini app. Impact on Drug Discovery and AI Market The announcement of TxGemma has implications for the pharmaceutical industry, particularly in accelerating drug development. Companies like Isomorphic Labs and others are exploring AI's potential to revolutionize drug discovery. Despite reported clinical trial failures, there is strong enthusiasm from both big pharma and investors regarding AI's capabilities. The launch of TxGemma has positively influenced AI-related tokens, with significant trading activity observed for Fetch.AI (FET) and SingularityNET (AGIX). Increased trading volumes and price surges indicate a favorable market sentiment towards AI developments, especially those applicable in healthcare. For enterprises looking to implement secure authentication solutions, SSOJet offers an API-first platform featuring SSO, MFA, and Passkey solutions, ensuring rob

Mar 31, 2025 - 09:20

Google DeepMind Unveils TxGemma: New Open AI for Drug Discovery

Originally published at ssojet

Google DeepMind has launched TxGemma, an open collection of AI models aimed at improving drug discovery and clinical trial predictions. Built on the Gemma model family, TxGemma is designed to streamline the drug development process and accelerate the identification of new treatments.

The TxGemma model addresses the significant challenges of the drug discovery process, where 90% of drug candidates fail to progress beyond phase 1 trials. By utilizing large language models (LLMs), TxGemma enhances the prediction of therapeutic properties throughout the research pipeline. Researchers can identify promising drug targets and assess clinical trial outcomes more effectively.

Jeremy Prasetyo, CEO of TRUSTBYTES, mentioned, "AI that explains its own predictions is a game-changer for drug discovery—faster insights mean faster breakthroughs in patient care." TxGemma, as a successor to Tx-LLM, expands its capabilities and is available as an open-source model.

Image courtesy of Blockchain.News

The model is trained on 7 million examples and comes in three sizes—2B, 9B, and 27B parameters. It specializes in:

Classification: Predicting molecule permeability across the blood-brain barrier.
Regression: Estimating drug binding affinity.
Generation: Inferring reactants from chemical reactions.

In benchmark tests, the 27B Predict model outperformed or matched specialized models on 64 of 66 key tasks. For further details, refer to the published paper.

TxGemma-Chat allows for interactive AI experiences, enabling researchers to ask complex questions and receive detailed explanations. Researchers can fine-tune the model for specific needs using a Colab notebook.

Additionally, Google DeepMind is introducing Agentic-Tx, which integrates TxGemma into multi-step research workflows. This tool combines TxGemma with Gemini 2.0 Pro and utilizes 18 specialized tools to enhance research capabilities, as demonstrated in benchmarks like Humanity’s Last Exam and ChemBench.

TxGemma is available on Vertex AI Model Garden and Hugging Face, which allows researchers to experiment with the models and use fine-tuning tools.

Introduction of Gemma 3

The Gemma family has expanded with the introduction of Gemma 3, which is described as the most capable, portable, and responsible open model yet. Celebrating over 100 million downloads, the Gemma models have fostered a vibrant community with over 60,000 variants.

Gemma 3 is designed for high performance on single GPU or TPU hosts, with several key capabilities:

Outperforming competitors like Llama3-405B and DeepSeek-V3, it provides state-of-the-art performance for its size.
Supporting over 35 languages out-of-the-box, with pre-trained support for up to 140 languages.
Enhanced text and visual reasoning capabilities for applications analyzing images, text, and videos.
A 128k-token context window for processing large volumes of information.
Function calling capabilities to automate tasks and enhance agentic experiences.

Developers can seamlessly integrate Gemma 3 into their existing workflows using tools like Hugging Face Transformers, Ollama, and Google AI Edge.

Gemma 3 and its safety counterpart, ShieldGemma 2, offer built-in safety measures for image applications, categorizing outputs across three safety areas: dangerous content, sexually explicit content, and violence.

Gemini 2.5 Model Launch

Gemini 2.5, described as Google's most intelligent AI model, enhances reasoning capabilities. The first release, Gemini 2.5 Pro Experimental, leads benchmarks significantly, showcasing strong reasoning and coding abilities.

Image courtesy of Google AI

Gemini 2.5 models can analyze information, draw logical conclusions, and make informed decisions. The new model features:

Advanced reasoning capabilities that surpass previous versions.
A 1 million token context window, with plans for a 2 million token window.
Enhanced performance across math, coding, and science benchmarks.

Developers can start experimenting with Gemini 2.5 Pro in Google AI Studio or through the Gemini app.

Impact on Drug Discovery and AI Market

The announcement of TxGemma has implications for the pharmaceutical industry, particularly in accelerating drug development. Companies like Isomorphic Labs and others are exploring AI's potential to revolutionize drug discovery. Despite reported clinical trial failures, there is strong enthusiasm from both big pharma and investors regarding AI's capabilities.

The launch of TxGemma has positively influenced AI-related tokens, with significant trading activity observed for Fetch.AI (FET) and SingularityNET (AGIX). Increased trading volumes and price surges indicate a favorable market sentiment towards AI developments, especially those applicable in healthcare.

For enterprises looking to implement secure authentication solutions, SSOJet offers an API-first platform featuring SSO, MFA, and Passkey solutions, ensuring robust user management and directory synchronization. Explore our offerings to enhance your enterprise security today.